Skip to content

Feasible Region

Turn an optimization problem's list of constraints into a single geometric set — the intersection of everything they permit — whose shape (empty, convex, bounded, its boundary) governs whether a solution exists, where it sits, and which methods will find it.

Core Idea

A feasible region is the set of all points in a candidate space that simultaneously satisfy every constraint imposed on an optimisation problem — geometrically, the intersection of the half-spaces, hypersurfaces, or other constraint sets cut out by each individual restriction. When constraints are expressed as inequalities or equalities over a real-valued variable space (typically ℝⁿ), each constraint carves out a permitted subset of that space, and the feasible region is what remains after all restrictions are applied jointly. Optimisation then asks which point of this intersection extremises an objective function; the feasible region is the domain on which the search takes place, and its geometric properties govern which algorithms are applicable and what guarantees can be made about the existence and location of optima.

The structural move the feasible region executes is the conversion of a list of rules into a single geometric object: many separate constraint specifications become one set with shape, dimension, vertices, boundary, and topological properties that can be reasoned about in aggregate. In linear programming, each linear inequality defines a half-space, and the feasible region is a convex polytope — a bounded or unbounded polyhedron with finitely many vertices. The fundamental theorem of linear programming states that if an optimal solution exists, it is attained at a vertex of this polytope, which is the structural fact underlying the simplex method: optimisation over a convex polytope reduces to checking finitely many vertices. The feasibility question — whether the intersection is non-empty — must be settled before any search for an optimum begins; an infeasible set of constraints means no solution exists, regardless of the objective. When the feasible region is convex (as in linear and convex programmes), every local optimum is a global optimum, and gradient-based algorithms converge to the global solution without being trapped in inferior local minima. When it is non-convex (as in integer programmes, mixed-integer programmes, and many engineering design problems), the geometry is far more complex — the set of feasible points may be disconnected, its interior may be empty (as in pure integer programmes, where the feasible region is a discrete lattice), and finding even a feasible solution, let alone an optimal one, can be NP-hard. In the Karush–Kuhn–Tucker conditions, the boundary of the feasible region plays a specific role: constraints that are active at the optimum (equality holding) contribute to the stationarity conditions through Lagrange multipliers, while inactive constraints (strict inequality) do not — the geometric concept of active versus inactive constraint is exactly the distinction between being on the boundary versus in the interior of the feasible region.

Structural Signature

Sig role-phrases:

  • the candidate space — the ambient set over which the search ranges (typically ℝⁿ, sometimes a discrete structure)
  • the constraints — the individual restrictions (inequalities, equalities) that each carve out a permitted subset of the candidate space
  • the intersection — the feasible region itself: the single geometric object remaining after all per-constraint subsets are intersected, the move from rule-list to one set with a shape
  • the emptiness check — the prior feasibility question: whether the intersection is non-empty, which must be settled before any optimum is sought (empty means no answer regardless of objective)
  • the convexity property — the regularity deciding tractability: when convex, every local optimum is global and gradient methods converge; when non-convex (disconnected lattice, empty interior) even finding a feasible point can be NP-hard
  • the boundedness-and-polyhedrality — boundedness (with a continuous objective) guaranteeing an optimum exists, and polyhedrality placing any optimum at a vertex (fundamental theorem of LP), reducing an uncountable search to finitely many corners (simplex)
  • the boundary-versus-interior structure — active constraints (equality holding, on the boundary) binding and carrying Lagrange multipliers / shadow prices (KKT), inactive ones (strict inequality, in the interior) loosenable for free
  • the non-convex limitation — the characteristic boundary: where the region is non-convex, disconnected, or discrete, local-equals-global fails, vertices may miss the optimum, and the convex read-offs must give way to global or combinatorial methods

What It Is Not

  • Not the optimal solution. The feasible region is the domain the search ranges over — every point satisfying all constraints — not the best point in it. Optimisation then extremises an objective over this set; the region defines where the search happens and what guarantees apply, but it is not itself the answer.
  • Not a single constraint. A constraint is one rule cutting out a permitted subset; the feasible region is the intersection of all such subsets — the single geometric object remaining after every restriction is applied jointly. The whole move is from a list of rules to one set with a shape, dimension, and boundary.
  • Not guaranteed non-empty. The intersection can be empty, in which case the problem has no answer regardless of the objective. Feasibility — does any point satisfy all constraints at once? — must be settled before optimality is even attempted; mistaking an infeasibility ("this cannot be done at all") for a poor optimum ("this is the best we can do") is exactly the category error the concept forecloses.
  • Not the state space. The state space is the ambient set of all configurations a system can occupy, often unconstrained; the feasible region is the constrained subset of that space carved out by the restrictions. Searching the full state space without intersecting the constraints is searching the wrong set.
  • Not the constraint prime itself. The feasible region is the geometric visualization of constraint conjunction in a coordinate space — load-bearing inside optimisation, but adding no substrate-independent skeleton beyond constraint. Its distinctive cargo (the vertex theorem, simplex, KKT Lagrange multipliers, NP-hardness of non-convex feasibility) is optimisation-internal; cross-domain, the transferable insights are inherited from constraint + convexity + optimization, not from "feasible region."

Scope of Application

Because a feasible region is a mathematical construct — the conjunction of constraints into an admissible set in a coordinate space — not a causal mechanism, it applies literally wherever that precondition holds: a coordinate space with restrictions intersected over it. The fields below are genuine uses of the same construct; where the term is borrowed outside a real coordinate-space-with-constraints setting it is the image of constraint conjunction, and the looser borrowings belong to the parent constraint + convexity + optimization.

  • Mathematical optimization — the home object: linear-programming polytopes, convex-programming admissible sets, semidefinite cones, and integer-programming lattices, read by settling emptiness, classifying convexity for tractability, exploiting polyhedrality for vertex optima (simplex), and locating active constraints (KKT).
  • Engineering design — the design space: the set of designs meeting load, yield, clearance, and weight limits jointly.
  • Economics — admissible bundles and outputs: budget sets, production-possibility frontiers, and technological-feasibility sets.
  • Operations research — combinatorial admissibility: scheduling polytopes and network-flow feasibility sets.
  • Control theory — admissible states and inputs: reachable, invariant, and admissible-control sets.
  • Robotics — collision-free configurations: configuration space minus obstacle regions ("C-free") for motion planning.
  • Public policy and decision analysis — the permitted-action set: the actions remaining after legal and resource constraints intersect.

Clarity

The decisive clarity the feasible region buys is the separation of feasibility from optimality — two questions that intuition collapses into a single "what's the best choice?" Naming the region makes it impossible to skip the prior question: does any point satisfy all the constraints at once? An infeasible problem has no answer regardless of how good the objective is, so emptiness of the intersection must be settled before the search for an optimum even begins. This dissolves a common category error in applied work — demanding the optimal use of an instrument whose constraints cannot be jointly met, treating an infeasibility as if it were merely a poor optimum. Once the distinction is in hand, "this is the best we can do" and "this cannot be done at all" are recognised as different failures requiring different remedies.

The concept's second clarifying move is the conversion of a list of rules into a single object with a shape. Many separate inequalities — a flour budget, an oven-time limit, non-negativity — stop being a checklist and become one set whose geometry can be reasoned about in aggregate: convex or not, bounded or not, with vertices, a boundary, an interior. That relocation from rule-by-rule checking to shape-level reasoning is what makes the deepest questions in optimisation askable. Is the region convex? becomes the question that decides whether a local optimum can be trusted as global, and therefore whether gradient descent can be trapped in an inferior basin. Where does the optimum live — interior or boundary? becomes the active-versus-inactive-constraint distinction that the KKT conditions turn on, since only constraints that bind at the solution carry Lagrange multipliers. The framing thereby tells the practitioner exactly which constraints can be loosened for free (the inactive ones, perturbable without effect) and which set the price of the solution (the active ones on the boundary) — the geometric content behind shadow prices and sensitivity analysis. The sharper question is no longer "what is the best point?" but "what is the shape of the set I am searching, and does that shape guarantee my method will find the answer?"

Manages Complexity

An optimisation problem arrives as a long, unstructured list — a budget here, a capacity limit there, a non-negativity requirement, a logical exclusion — and the naive way to handle a candidate point is to walk the list and check every rule, repeating that for every point the search visits. Across thousands of constraints and an uncountable candidate space this rule-by-rule bookkeeping is both endless and uninformative: it never tells the analyst anything about the problem as a whole. The feasible region collapses that list into a single geometric object, the intersection of the per-constraint sets, and shifts the analyst's attention from checking points to reasoning about a shape — its convexity, boundedness, dimension, vertices, boundary, interior, and whether it is empty at all. The sprawl of separate rules becomes one set with aggregate properties, and a handful of those properties is what the practitioner now tracks, reading off from them the guarantees that would otherwise have to be re-established problem by problem.

The compression pays off because the shape's properties partition optimisation problems into qualitatively distinct regimes, each with its own verdict that follows without solving the problem afresh. The first and most basic branch is emptiness: if the intersection is empty the problem has no answer whatever the objective, so feasibility is settled before optimality is even attempted — which is exactly the category error the shape-level view forecloses, refusing to mistake an infeasibility for a poor optimum. The second branch is convexity: when the region is convex (linear and convex programmes) every local optimum is global and gradient methods converge to the true answer, whereas when it is non-convex (integer and mixed-integer programmes, where the region may be a disconnected lattice with empty interior) even finding a feasible point can be NP-hard — so a single geometric property tells the analyst in advance whether the search is tractable or potentially intractable, and whether a local solution can be trusted. The third branch is boundedness, which under continuity of the objective guarantees an optimum exists, and polyhedrality, which (by the fundamental theorem of linear programming) places any optimum at a vertex and so reduces an uncountable search to checking finitely many corners — the structural fact the simplex method exploits. A fourth read-off concerns which constraints matter: the active constraints on the boundary set the price of the solution and carry the Lagrange multipliers of the KKT conditions, while the inactive ones in the interior can be loosened for free — the geometric content behind shadow prices and sensitivity analysis. So the move is from an open-ended checklist of rules to one set whose shape is the compressed carrier of the problem's character, with a branch structure — empty/non-empty, convex/non-convex, bounded/unbounded, active/inactive — from which solvability, the right algorithm, the existence and location of optima, and the sensitivity of the solution are all read off.

Abstract Reasoning

The feasible region licenses a set of moves on any optimisation problem, all routed through the conversion of a constraint list into a single geometric shape and the properties of that shape. Boundary-drawing (the founding move) — settle feasibility before optimality: the foundational move is to ask, before searching for the best point, whether any point satisfies all constraints at once — whether the intersection is non-empty — because an infeasible problem has no answer regardless of the objective. The analyst reasons from "is this set of constraints jointly satisfiable?" to "if the intersection is empty, stop — there is nothing to optimise," so the move is to refuse the common category error of treating an infeasibility as a merely poor optimum, recognising "this cannot be done at all" and "this is the best we can do" as different failures with different remedies. Diagnostic — reason about a shape, not a checklist: the move is to collapse the separate inequalities and equalities into one set with aggregate geometric properties — convexity, boundedness, dimension, vertices, boundary, interior — and reason about the shape rather than walk the list point by point. The analyst reasons from "a flour budget, an oven-time limit, a non-negativity rule" to "one polytope whose shape carries the problem's character," so the move is to relocate from rule-by-rule checking to shape-level reasoning that yields guarantees about the problem as a whole. Predictive (the signature move) — read tractability and the trustworthiness of a local optimum off convexity: the decisive move is to ask "is the region convex?" because that single property decides whether a local optimum can be trusted as global and whether gradient methods can be trapped. The analyst reasons from "the region is convex (linear or convex programme)" to "every local optimum is global, and gradient descent converges to the true answer," and from "the region is non-convex (integer or mixed-integer programme, possibly a disconnected lattice with empty interior)" to "even finding a feasible point can be NP-hard, and a local solution cannot be trusted" — so a geometric property predicts in advance whether the search is tractable. Predictive — read existence and the location of optima off boundedness and polyhedrality: the move is to use boundedness (under a continuous objective, an optimum exists) and polyhedrality (by the fundamental theorem of linear programming, any optimum sits at a vertex) to convert an uncountable search into a finite one. The analyst reasons from "the feasible polytope is bounded with finitely many vertices" to "check the corners — the optimum is among them," the structural fact the simplex method exploits, so existence and location are read off the shape rather than established by exhaustive search. Diagnostic / interventionist — locate the optimum on the boundary versus interior to find which constraints bind: the move is to ask whether the optimum lies on the boundary or in the interior, because active constraints (equality holding, on the boundary) set the price of the solution and carry the Lagrange multipliers of the KKT conditions, while inactive ones (strict inequality, in the interior) can be loosened for free. The analyst reasons from "this constraint is active at the optimum" to "it binds, carries a shadow price, and tightening or loosening it moves the objective," and from "this one is inactive" to "it can be perturbed without effect" — the geometric content behind sensitivity analysis, telling the practitioner exactly which constraints to renegotiate. The boundary on every move is the geometry the read-offs assume: the convexity guarantee, the vertex theorem, and the active/inactive analysis hold for the constraint types that produce well-behaved shapes (linear half-spaces, convex sets), so the move where the region is non-convex, disconnected, or discrete is to recognise that local-equals-global fails, vertices may not capture the optimum, and the problem may be intractable — switching to global or combinatorial methods rather than trusting the convex read-offs.

Knowledge Transfer

Within mathematical optimisation the feasible region transfers as mechanism, and its full apparatus — the feasibility-before-optimality discipline, the rule-list-to-shape conversion, the convex/non-convex tractability split, boundedness-and-polyhedrality giving vertex optima (the fundamental theorem of LP, the simplex method), and the active/inactive-constraint analysis with its Lagrange multipliers and shadow prices (KKT) — carries across the subfields without translation. Linear-programming polytopes, convex-programming admissible sets, semidefinite cones, and integer-programming lattices are all read the same way: settle emptiness first, classify by convexity to predict tractability, exploit polyhedrality to reduce an uncountable search to finitely many corners, and locate the optimum on the boundary to see which constraints bind. The geometric vocabulary, the branch structure, and the algorithm-selection read-offs all move within optimisation theory intact.

Beyond optimisation the cross-domain travel is genuine but is best read as case (B): the picture appears across engineering (design spaces under load, yield, clearance, and weight limits), economics (budget sets, production-possibility frontiers, technological-feasibility sets), operations research (scheduling polytopes, network-flow feasibility), control theory (reachable, invariant, and admissible-control sets), robotics (configuration space minus collision regions, "C-free," for motion planning), and public policy (the set of permitted actions after legal constraints intersect). In every one of these, what actually recurs is not the optimization-specific machinery but the parent the feasible region instantiates — the conjunction of restrictions into an admissible set — and that pattern is already carried by constraint (the rule that cuts a permitted subset), convexity / convex_set (the regularity that makes the intersection tractable), optimization (the search the set serves), and tradeoff / pareto_frontier (the boundary of the jointly achievable). The feasible region is the geometric visualization of constraint conjunction in a coordinate space — load-bearing inside optimisation, but adding no substrate-independent skeleton of its own beyond constraint. So the honest cross-domain lesson is to carry those parents: the transferable insights an economist, roboticist, engineer, or policy analyst actually uses — constraint conjunction shrinks the admissible set, emptiness means "cannot be done at all" (not "a poor optimum"), only active constraints bind and carry a price — are inherited from constraint + convexity + optimization, not from "feasible region." Where the term is borrowed outside a genuine coordinate-space-with-constraints setting it is the image of constraint conjunction rather than the optimization construct, and the LP/KKT cargo (vertex theorem, simplex, Lagrange multipliers, NP-hardness of non-convex feasibility) stays home (see Structural Core vs. Domain Accent).

Examples

Canonical

Take the textbook two-variable linear program: maximize P = 3x + 4y subject to x + y ≤ 4, x + 3y ≤ 6, and x, y ≥ 0. Each inequality cuts a half-plane; their intersection is a bounded convex polygon with four corners — (0,0), (4,0), (0,2), and the point where x + y = 4 meets x + 3y = 6. Solving those two: subtracting gives 2y = 2, so y = 1 and x = 3, the vertex (3,1) (check: 3+1 = 4 ✓, 3 + 3 = 6 ✓). By the fundamental theorem of linear programming, the optimum sits at a vertex, so evaluate P at all four: 0, 12, 8, and 3(3)+4(1) = 13. The maximum is 13 at (3,1), where both constraints hold with equality. The uncountable search collapses to checking four corners — exactly what Dantzig's simplex method (1947) walks.

Mapped back: ℝ² with x, y ≥ 0 is the candidate space; the two inequalities are the constraints, and the four-cornered polygon is the intersection. The polygon being bounded and polyhedral is the boundedness-and-polyhedrality that guarantees the optimum lives at a vertex. At (3,1) both inequalities are active — the boundary-versus-interior structure — so both bind and carry shadow prices, while non-negativity is slack.

Applied / In Practice

George Stigler's 1945 "diet problem" is the historical field deployment. Stigler asked for the cheapest daily diet meeting a set of minimum nutrient requirements (calories, protein, calcium, vitamins, and others) drawn from a list of 77 foods. Each nutrient minimum is a linear constraint, and their intersection is the feasible set of food-quantity combinations that keep a person adequately nourished; minimizing cost over that set is a linear program. Lacking a solution method, Stigler guessed heuristically at an annual cost of about $39.93 (1939 prices). A few years later, once Dantzig's simplex method existed, the true optimum was computed as roughly $39.69 per year — remarkably close, and achieved with only a handful of the 77 foods entering at nonzero quantity, exactly as the vertex theorem predicts.

Mapped back: The nonnegative food-quantity vectors form the candidate space; each nutrient minimum is one of the constraints, and adequate diets are the intersection. That the intersection is nonempty passes the emptiness check, and its convex polyhedral form makes the LP tractable and puts the optimum at a vertex — which is why only a few foods appear at nonzero amounts, the geometric signature of boundedness-and-polyhedrality.

Structural Tensions

T1: Feasibility-before-optimality as a cheap prior question versus feasibility as the hard problem (the discipline collapses in the non-convex regime). The concept's founding discipline is to settle emptiness first, because an infeasible problem has no answer regardless of the objective — a logically prior, seemingly cheap gate. But that priority holds cleanly only where the region is well-behaved: for a non-convex or integer feasible set, finding even a single feasible point can itself be NP-hard, exactly as hard as the optimisation it was supposed to precede. So the tidy sequencing — first check the intersection is non-empty, then search it — silently assumes the tractable regime, and in the hard regime the "prior" question is not prior in cost at all. The gate that forecloses the infeasibility-versus-poor-optimum category error can be as expensive as the whole problem. Diagnostic: Is confirming non-emptiness a quick feasibility check here, or is it, for this constraint type, an NP-hard search dressed as a preliminary?

T2: Convexity as a single tractability oracle versus the cost and fragility of establishing it. One geometric property — convexity — decides everything downstream: whether a local optimum can be trusted as global, whether gradient methods converge, whether the search is tractable. That compression is the concept's most powerful read-off. But it presupposes you already know the region is convex, and convexity is a property of the constraint forms, not something visible on the surface of a rule list; a single non-convex constraint, or an integrality requirement, flips the whole verdict from "local equals global" to "possibly NP-hard." The property that carries all the guarantees is therefore both expensive to certify in general and brittle to small changes in problem statement. Trusting the convex read-offs without verifying convexity imports guarantees the shape may not actually possess. Diagnostic: Has the region's convexity been established from the constraint structure, or merely assumed because the two-variable picture looked like a tidy polygon?

T3: The vertex theorem's finite reduction versus its exponential blow-up (finite is not tractable, and low-dimensional geometry misleads). The fundamental theorem of LP converts an uncountable search into checking finitely many corners — the structural fact the simplex method exploits, and the source of the concept's most satisfying "reduce to the vertices" move. Yet a polytope in n dimensions can carry a number of vertices that grows exponentially in the constraints, so "finite" and "tractable" part ways: enumerating corners is hopeless in high dimension, and the simplex walk can in the worst case visit exponentially many. Worse, the shape-level intuition the concept sells is trained on visualizable 2-D and 3-D polygons, where boundary, interior, and vertices are legible — precisely the intuitions that fail when volume concentrates near the boundary in high dimension. The geometric picture that makes the concept teachable is the one that misleads at scale. Diagnostic: Is "check the vertices" a small finite enumeration here, or an exponential corner count where the finiteness guarantee buys no tractability?

T4: Active-versus-inactive as a clean free/binding split versus its purely local validity (shadow prices expire under real change). The boundary-versus-interior read-off tells the practitioner which constraints set the price (active, carrying Lagrange multipliers) and which can be loosened for free (inactive, in the interior) — the geometric content behind sensitivity analysis. But "loosen for free" and the shadow price attached to a binding constraint are local facts, valid only within a range: perturb a right-hand side far enough and a slack constraint becomes active, the optimal vertex jumps to a different corner, and the shadow price that seemed to quantify a constraint's worth is superseded. The split reads as a stable classification of constraints into "matters" and "doesn't," when it is really a snapshot of which constraints bind at this optimum for this data. Treating the free/binding partition as durable renegotiation guidance overreaches its local warranty. Diagnostic: Is the inactive constraint being loosened within the range where it stays slack, or by an amount large enough to make it bind and relocate the optimum?

T5: Optimality on the boundary versus robustness (the best point is the most fragile point). For linear and most convex programmes the optimum lives on the boundary, pinned where active constraints meet — the vertex the simplex method finds. That location is what makes it optimal and, by the same fact, what makes it fragile: an optimum defined by binding constraints sits exactly where the smallest tightening of a constraint, or a thin sliver of feasible room, pushes it to infeasibility or a distant corner. A feasible region that is non-empty but nearly degenerate passes the emptiness check yet yields a solution that is mathematically valid and practically unrealizable, and a region whose constraints drift over time leaves yesterday's boundary optimum stranded. The concept's own logic — extremise on the admissible set — drives the answer to the least robust part of the set. Diagnostic: Does the chosen point buy its optimality by sitting hard against active constraints, and how much do those constraints have to move before it is no longer feasible or no longer best?

T6: Autonomy versus reduction (its own optimisation construct or the geometric picture of constraint/convexity/optimization). "Feasible region" carries real optimisation-internal cargo — the vertex theorem, simplex, the KKT active/inactive analysis with Lagrange multipliers and shadow prices, the NP-hardness of non-convex feasibility — and within mathematical optimisation it travels intact as mechanism across LP polytopes, convex sets, semidefinite cones, and integer lattices. But the entry is unusually candid that beyond optimisation it adds no substrate-independent skeleton of its own: what recurs across engineering design spaces, budget sets, C-free configuration spaces, and permitted-action sets is the parent — constraint conjunction into an admissible set — already housed in constraint, convexity/convex_set, optimization, and tradeoff/pareto_frontier. The feasible region is the geometric visualization of constraint conjunction in a coordinate space, not an independent cross-domain prime. The tension is between a load-bearing named construct inside optimisation and a mere image of its parents outside it. Diagnostic: Resolve toward the parents (constraint + convexity + optimization) when asking what an economist or roboticist actually reuses; toward named feasible region when the setting is a genuine coordinate space with intersected constraints and the LP/KKT cargo applies.

Structural–Framed Character

The feasible region sits at mixed-structural on the structural–framed spectrum — a formal mathematical construct with strong structural credentials, held back from the pole not by domain-embeddedness of a natural mechanism (as with isostasy) but by the unusual fact that its distinctive content is discipline-internal apparatus while its portable content is wholly inherited from its parents. The criteria split cleanly. On evaluative weight it is decisively structural: a feasible region is neither good nor bad, praises and blames nothing — it is a geometric set whose shape carries verdicts about existence and tractability, not a normative classification like the fallacies elsewhere in this batch. On institutional origin it is structural: the region is a mathematical fact about the intersection of constraint sets in a coordinate space, not an artifact of a survey, agency, or convention — Dantzig and Stigler exploited and named geometry that the constraints already determine. And on human-practice-bound it patterns structural in the relevant sense: it is not constituted by any social or normative practice that would dissolve if the practice were removed; it is a representational construct that holds literally wherever its precondition holds — a candidate space with restrictions intersected over it — with the same geometry obtaining whether or not anyone reasons about it.

What keeps it off the structural pole is the pair vocab-travels and import-vs-recognize, and the entry is candid about why. Its distinctive vocabulary — vertex theorem, simplex, KKT active/inactive constraints, Lagrange multipliers and shadow prices, polyhedrality, NP-hardness of non-convex feasibility — is irreducibly optimisation-internal and does not float free of a coordinate-space-with-constraints setting; carry the concept to an economist's budget set or a roboticist's C-free space and that apparatus stays home. Within optimisation the mechanism transfers as recognition across LP polytopes, convex sets, semidefinite cones, and integer lattices; beyond it, what actually recurs is not "feasible region" but the parent it visualizes, so the reuse is inheritance-from-parents rather than the construct itself travelling.

The portable structural skeleton is constraint conjunction into an admissible set — many restrictions intersected to a single permitted subset whose emptiness, convexity, and boundary govern what can be found in it. That skeleton is genuinely substrate-portable, which is what tempts a structural reading and gives the concept its high placement. But it does not lift the feasible region to a prime, because — as the entry states outright — the concept adds no substrate-independent skeleton of its own beyond constraint: it is the geometric picture of constraint conjunction that the catalog already carries as constraint, convexity/convex_set, optimization, and tradeoff/pareto_frontier. The cross-domain reach belongs to those parents, while the LP/KKT/simplex cargo is the domain-accented expression that stays in optimisation. Its character: structurally neutral and formally portable in skeleton — a real, evaluatively neutral geometric visualization of constraint conjunction — but distinctive only in its optimisation-internal apparatus, so that everything that travels is inherited from its parents and everything proprietary stays home, leaving it mixed-structural rather than a free-floating prime.

Structural Core vs. Domain Accent

This section decides why the feasible region is a domain-specific abstraction and not a prime — and here the case is unusually clean, because the concept is candid that its portable content is entirely inherited and its distinctive content is entirely discipline-internal.

What is skeletal (could lift toward a cross-domain prime). Strip the optimisation apparatus and a real relational structure survives: many separate restrictions are conjoined — intersected — into a single admissible set, whose emptiness, convexity, and boundary govern what can be found within it. The portable pieces are abstract — a candidate space, a family of rules each cutting a permitted subset, their intersection into one set, and the shape-level properties (non-empty or not, convex or not, where the boundary sits) that carry verdicts about existence and tractability. That skeleton is genuinely substrate-portable: it recurs in engineering design spaces under load and weight limits, economic budget sets and production-possibility frontiers, robotics' collision-free configuration space, and the permitted-action set left after legal constraints intersect. Precisely because it recurs, it is carried by the parents the entry instantiates — constraint (the rule that cuts a permitted subset), convexity/convex_set (the regularity that makes the intersection tractable), optimization (the search the set serves), and tradeoff/pareto_frontier (the boundary of the jointly achievable). But that is the core the feasible region shares — and, unusually, it is nearly all of what the concept portably contains.

What is domain-bound. What makes this specifically the feasible region is optimisation-internal apparatus, and it is exactly this that does not survive extraction: the fundamental theorem of LP placing any optimum at a vertex of a bounded polytope, the simplex method that walks those corners, the KKT active-versus-inactive-constraint analysis with its Lagrange multipliers and shadow prices, the convex/non-convex tractability split, and the NP-hardness of non-convex feasibility on a disconnected discrete lattice. This vocabulary — polyhedrality, vertices, boundary versus interior, semidefinite cones, integer lattices — is the worked language of mathematical optimisation, and the empirical cases (the two-variable LP polygon, Stigler's diet problem) are drawn from it. The decisive test: remove the coordinate-space-with-intersected-constraints setting — borrow "feasible region" for a situation with no real coordinate space and no metric geometry — and the LP/KKT cargo has nothing to attach to; what remains is merely the image of constraint conjunction, which is the parent constraint, not this construct.

Why this does not clear the prime bar. A prime is a relational structure whose vocabulary travels and whose cross-domain transfer is recognition of the same mechanism, not analogy — and the feasible region is the rare entry that fails the bar not by carrying domain baggage that swamps a portable core, but by adding no substrate-independent skeleton of its own beyond constraint. Its transfer is bimodal. Within mathematical optimisation it moves intact as mechanism — the feasibility-before-optimality discipline, the rule-list-to-shape conversion, the convex/non-convex split, the vertex theorem and simplex, and the active/inactive analysis all carry without translation across LP polytopes, convex sets, semidefinite cones, and integer lattices. Beyond optimisation what recurs is not "feasible region" but the parent it visualizes: the insights an economist, roboticist, engineer, or policy analyst actually reuses — constraint conjunction shrinks the admissible set, emptiness means "cannot be done at all" rather than "a poor optimum," only active constraints bind and carry a price — are inherited from constraint + convexity + optimization, not from this construct. So when the bare structural lesson is needed cross-domain, it is already carried, in more general form, by those parents. The cross-domain reach belongs to them; "feasible region," as named, is the geometric visualization of constraint conjunction whose distinctive LP/KKT apparatus is load-bearing inside optimisation and stays home.

Relationships to Other Abstractions

Local relationship map for Feasible RegionParents appear above the current abstraction, mutual partners to the right, and children below. Node labels state whether each abstraction is prime or domain-specific; colors identify relation types.Feasible RegionDOMAINPrime abstraction: Constraint — is a decomposition ofConstraintPRIMEPrime abstraction: Intersection — is a kind ofIntersectionPRIME

Current abstraction Feasible Region Domain-specific

Parents (2) — more general patterns this builds on

  • Feasible Region is a kind of Intersection Prime

    A Feasible Region is the Intersection specialized to the permitted sets carved out by every constraint over one candidate space.

  • Feasible Region is a decomposition of Constraint Prime

    Removing optimization geometry from Feasible Region leaves Constraint's exact candidate-domain, admissibility-condition, and feasible-set structure.

Hierarchy paths (2) — routes to 2 parentless roots

Not to Be Confused With

  • The objective function. The scalar map that assigns each candidate point a value to be extremised. The feasible region is the domain the objective is optimised over; the objective is what ranks the points within it — and the two are orthogonal inputs, since swapping the objective leaves the region unchanged while changing a constraint moves the region and leaves the objective standing. Tell: does the item assign points a value to be maximised or minimised (objective function), or delimit which points are admissible at all (feasible region)?

  • A convex set / the convex hull. Convexity is a property a feasible region may or may not have — the property that decides local-equals-global and tractability — and the convex hull is the smallest convex set containing a given point set, a different construction entirely. A feasible region need not be convex: integer programmes give a disconnected discrete lattice with empty interior. Tell: convexity or the hull names a regularity (or a set built to have it), whereas the feasible region is the constraint intersection itself, whose convexity is a separate question the analyst must check.

  • A polytope / polyhedron. The bounded convex shape a linear feasible region takes — a polyhedron with finitely many vertices, the object the fundamental theorem of LP and the simplex method exploit. But the feasible region is the general intersection under any constraint types; only in linear programming is it guaranteed polyhedral, and integer or non-convex constraints break the polytope picture. Tell: a polytope is the special linear-programming form; the feasible region is the general admissible set whatever its geometry, polyhedral or not.

  • The Pareto (efficient) frontier. The set of non-dominated points in objective/criterion space, tracing the boundary of jointly-achievable performance across competing objectives. The feasible region lives in decision-variable space and is cut out by constraints, not by dominance among objectives; the Pareto frontier is one of the parents the entry instances for the boundary of the jointly achievable. Tell: is the set drawn in the space of outcomes and defined by non-dominance (Pareto frontier), or in the space of decision variables and defined by intersected constraints (feasible region)?

  • A constraint-satisfaction problem (feasibility alone). A CSP asks only whether some point satisfies every constraint — the emptiness check standing by itself, with no objective to extremise. The feasible region serves an optimisation: the same admissible set, now searched for its best point under an objective. Tell: is the goal merely to find one admissible point or prove non-emptiness (a CSP — the feasibility question by itself), or to find the best admissible point under an objective (optimisation over the feasible region)?

  • The constraint / convexity / optimization parents (the umbrella it instances). The substrate-neutral relational primes — a rule cutting a permitted subset, the regularity that makes an intersection tractable, the search the set serves — that the feasible region is the coordinate-space geometric visualisation of, not a peer to be sorted against. Tell: strip the coordinate space and the LP/KKT apparatus, and what actually recurs cross-domain (constraint conjunction shrinks the admissible set; emptiness means "cannot be done at all"; only active constraints bind) is these parents at work, not "feasible region." (Treated fully in the sections above.)

Neighborhood in Abstraction Space

Feasible Region sits in a sparse region of the domain-specific corpus (75th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (309 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12